Bogie framework fault early warning method and system based on dynamic strain signals

Abstract

The invention discloses a bogie framework fault early warning method and system based on a dynamic strain signal, and relates to the technical field of bogie framework fault early warning, and the method comprises the steps: obtaining a real-time dynamic strain signal of train operation, and inputting the preprocessed real-time dynamic strain signal into a bogie framework fault early warning model, outputting a healthy dynamic strain data prediction value at the next moment through a healthy dynamic strain data prediction model in the model, and determining a dynamic strain data residual value based on the actual dynamic strain data value and the healthy dynamic strain data prediction value at the next moment; comparing the dynamic strain data residual error value with a dynamic residual error early warning threshold value and a static early warning threshold value, and performing fault early warning of different levels according to a comparison result; according to the method, historical health dynamic strain signals of train operation are utilized to train a bogie framework fault early warning model based on a health dynamic strain data prediction model and a dynamic early warning threshold generation module. According to the invention, more accurate online fault early warning can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of bogie structure fault early warning, and in particular to a bogie structure fault early warning method and system based on dynamic strain signals. Background Art

[0002] The bogie is an important load-bearing component of train locomotives and its structural safety is of vital importance. The bogie frame is the key to supporting the bogie. If the frame fails, it may cause instability during train operation, increase the friction and energy consumption of the train, affect the performance of the entire locomotive, and cause economic losses or even casualties. Therefore, in order to improve the reliability and operation safety of train locomotives and reduce maintenance costs, it is of great practical significance and practical need to carry out fault warning research for bogie frame failures that may occur during train operation.

[0003] In recent years, with the improvement of sensor accuracy and the rapid development of big data algorithms, data-driven methods based on deep learning feature extraction have gained increasing attention due to their advantages such as being able to discover complex structures and learn powerful feature representations from raw signals. However, data-driven methods based on deep learning still rely on a large amount of fault data for training, that is, to define, identify and diagnose fault features for known fault data. However, due to the different lifespans and costs of various mechanical structures, the fault data of many complex structures are difficult to obtain and the cost is high, making it impossible to train deep learning fault diagnosis models based on massive fault data. In other words, due to the scarcity of fault data, the bogie structure fault warning data model cannot be fully trained, the model performance is poor, and the fault warning task cannot be completed.

[0004] Since data-driven deep learning often has difficulty in achieving good training results due to the scarcity of fault signals, the selection of models trained with healthy signals can overcome this drawback. To this end, in view of the complex working environment of the bogie frame, labeled healthy data can be used based on machine learning. In the case of limited fault data resources, the health data characteristics can be deeply mined to construct a health data state space. On this basis, the existing health data in the source domain is used as a reference in this space to detect and analyze the rationality of the unlabeled data in the target domain. By clarifying and quantifying the differences between the two different data domains, and using custom fault thresholds to define warning signals, the accuracy of warnings can be greatly improved to avoid false alarms.

[0005] However, in practical applications, it is relatively easy to collect a source domain with a large fluctuation range of dynamic strain signals and a healthy label for the overall data, while it is relatively difficult to find a target domain with a similar fluctuation trend to the source domain dynamic strain signals. After clarifying the difference between healthy data and fault data, general machine learning methods lack appropriate quantitative standards in the process of quantifying the degree of difference, that is, they are unable to reasonably define fault data. In addition, in general fault warning models, the threshold for selecting diagnostic strategies is usually fixed, which is inconsistent with the actual complex and changeable working environment. Fault judgment based only on this static threshold will lead to a significant decrease in the accuracy of the diagnostic model warning signal. Summary of the invention

[0006] In order to address the deficiencies of the above-mentioned prior art, the present invention provides a bogie structure fault warning method and system based on dynamic strain signals, and applies the fault prediction model to the field of dynamic strain signal prediction and fault warning. When fault signals are scarce, only healthy dynamic strain signal data is used for model training without the need for a large amount of fault data. The healthy dynamic strain signal is used as a whole to construct a diagnostic model to achieve online fault warning. At the same time, a fault warning strategy is designed based on extreme value theory. Under complex working conditions, an adaptive dynamic threshold is established based on the dynamic strain residual, and the dynamic threshold is combined with the static threshold to improve the warning accuracy and effectively solve the fault warning problem of the bogie frame in the absence of fault data.

[0007] In a first aspect, the present disclosure provides a bogie structure fault early warning method based on dynamic strain signals.

[0008] A bogie frame fault early warning method based on dynamic strain signals, comprising:

[0009] Acquire the real-time dynamic strain signal of the train operation and perform data preprocessing on the acquired signal;

[0010] The preprocessed real-time dynamic strain signal is input into the bogie frame fault warning model. First, the healthy dynamic strain data prediction model in the model is used to output the predicted value of the healthy dynamic strain data at the next moment, and the dynamic strain data residual value is determined based on the actual dynamic strain data value and the predicted value of the healthy dynamic strain data at the next moment; then, the dynamic strain data residual value is compared with the dynamic residual warning threshold and the static warning threshold, and different levels of fault warning are performed according to the comparison results;

[0011] Among them, the historical healthy dynamic strain signals of train operation are used to train the bogie structure fault warning model based on the healthy dynamic strain data prediction model and the dynamic warning threshold generation module.

[0012] In a second aspect, the present disclosure provides a bogie structure fault warning system based on dynamic strain signals.

[0013] A bogie structure fault early warning system based on dynamic strain signals, comprising:

[0014] The data acquisition and preprocessing module is used to acquire the real-time dynamic strain signal of the train operation and perform data preprocessing on the acquired signal;

[0015] The fault warning module is used to input the preprocessed real-time dynamic strain signal into the bogie structure fault warning model, firstly output the predicted value of healthy dynamic strain data at the next moment through the healthy dynamic strain data prediction model in the model, and determine the residual value of dynamic strain data based on the actual dynamic strain data value and the predicted value of healthy dynamic strain data at the next moment; then compare the residual value of dynamic strain data with the dynamic residual warning threshold and the static warning threshold, and perform fault warnings of different levels according to the comparison results;

[0016] Among them, the historical healthy dynamic strain signals of train operation are used to train the bogie structure fault warning model based on the healthy dynamic strain data prediction model and the dynamic warning threshold generation module.

[0017] In a third aspect, the present disclosure further provides an electronic device, comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps of the method described in the first aspect are completed.

[0018] In a fourth aspect, the present disclosure further provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the method described in the first aspect.

[0019] One or more of the above technical solutions have the following beneficial effects:

[0020] 1. The present invention provides a bogie frame fault warning method based on dynamic strain signals, applies the fault prediction model to the field of dynamic strain signal prediction and fault warning, and in the case of scarce fault signals, only healthy dynamic strain signal data is used for model training without the need for a large amount of fault data; the healthy dynamic strain signal is used as a whole to construct a diagnostic model to achieve online fault warning, and at the same time, a fault warning strategy is designed based on extreme value theory. Under complex working conditions, an adaptive dynamic threshold is established based on the dynamic strain residual, the warning signal is dynamically defined, and the dynamic threshold is combined with the static threshold, so as to improve the diagnostic effect of the model, increase the warning credibility, and effectively solve the fault warning problem of the bogie frame in the case of scarce fault data.

[0021] 2. The present invention can realize the online monitoring and real-time early warning functions of dynamic strain data, and can avoid the limitation that traditional recurrent neural networks can only perform post-analysis. The construction of the healthy dynamic strain data prediction model adopts the echo state network to establish a normal or healthy dynamic strain state space model. Compared with the traditional recurrent neural network model, the network structure is simpler and the overfitting risk is smaller. The storage pool can filter out irrelevant noise signals of the input data, and it has dynamic characteristics and strong generalization ability.

[0022] 3. In this application, the training process of the fault warning model does not need to directly rely on fault data. Excellent training results can be achieved by relying solely on healthy dynamic strain signal data, which greatly reduces the dependence on scarce fault data and effectively reduces the overall cost of model development. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0024] Figure 1 It is a flow chart of the bogie structure fault early warning method based on dynamic strain signal according to an embodiment of the present invention;

[0025] Figure 2 It is a flowchart of a bogie structure fault early warning method based on dynamic strain signals according to an embodiment of the present invention;

[0026] Figure 3 It is a schematic diagram of the structure of a state echo network in an embodiment of the present invention;

[0027] Figure 4 Schematic diagram of the structure of a dynamic warning threshold generation module in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] It should be noted that the following detailed descriptions are exemplary only, are intended to describe specific embodiments, are intended to provide further explanation of the present invention, and are not intended to limit exemplary embodiments according to the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those of ordinary skill in the art to which the present invention belongs. In addition, it should also be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0029] Embodiment 1

[0030] This embodiment provides a bogie frame fault early warning method based on dynamic strain signals, such as Figure 1 As shown, the following steps are included:

[0031] Acquire the real-time dynamic strain signal of the train operation and perform data preprocessing on the acquired signal;

[0032] The preprocessed real-time dynamic strain signal is input into the bogie frame fault warning model. First, the healthy dynamic strain data prediction model in the model is used to output the predicted value of the healthy dynamic strain data at the next moment, and the dynamic strain data residual value is determined based on the actual dynamic strain data value and the predicted value of the healthy dynamic strain data at the next moment; then, the dynamic strain data residual value is compared with the dynamic residual warning threshold and the static warning threshold, and different levels of fault warning are performed according to the comparison results;

[0033] Among them, the historical healthy dynamic strain signals of train operation are used to train the bogie structure fault warning model based on the healthy dynamic strain data prediction model and the dynamic warning threshold generation module.

[0034] The bogie structure fault early warning method based on dynamic strain signal proposed in this embodiment is introduced in more detail through the following content.

[0035] First, the real-time dynamic strain signal of the train is obtained, and the obtained signal is preprocessed. Specifically, the optical fiber strain sensor is used to collect the real-time dynamic strain signal E of the bogie frame when the train is running at a sampling frequency of 1kHz. 1 ,e 2 ,e 3 ,…,e N ), the dynamic strain signal of the bogie frame during the current train operation is taken as the target domain data, where E represents a dynamic strain signal of the bogie frame, e i represents the i-th sampling point of the dynamic strain signal E, 1≤i≤N, and N is the total number of sampling points of the signal E. In this embodiment, a total of 1000 unlabeled dynamic strain signals are obtained in the target domain, where signals of different fault types can be constructed in a ratio of healthy: side beam fault: cross beam fault = 8:1:1.

[0036] Afterwards, the acquired dynamic strain signals are subjected to data preprocessing. That is, each dynamic strain signal is subjected to standardization processing. For each dynamic strain signal E=(e 1 ,e 2 ,e 3 ,…,e N ), the standardization formula is:

[0037]

[0038] Where ε represents the mean value of a dynamic strain signal E.

[0039] Secondly, the pre-processed real-time dynamic strain signal is input into the bogie frame fault warning model for fault warning.

[0040] In this embodiment, a bogie frame fault warning model is constructed and trained, and the fault warning model is used for fault judgment and warning. The bogie frame fault warning model consists of two parts: a healthy dynamic strain data prediction model based on an echo state network and a dynamic residual threshold generation module based on extreme value theory. The construction and training process specifically includes:

[0041] Step S1, obtaining historical healthy dynamic strain signals of train operation;

[0042] Step S2, performing data preprocessing on the acquired signal;

[0043] Step S3, based on the preprocessed historical healthy dynamic strain data, offline modeling of the bogie structure fault warning model is performed, including: first, constructing a healthy dynamic strain data prediction model based on the echo state network, and using the historical healthy dynamic strain data for training; secondly, using the trained healthy dynamic strain data prediction model to generate the residual value of the historical dynamic strain data, and using the residual value of the historical dynamic strain data to construct a dynamic warning threshold generation module based on extreme value theory.

[0044] In step S1, the historical healthy dynamic strain signal of the train operation is obtained. For the source domain, the optical fiber strain sensor is used to collect the normal and healthy bogie frame dynamic strain signal E of the train operation at a sampling frequency of 1kHz. 1 ,e 2 ,e 3 ,…,e N ), a total of 4000 labeled signals were obtained, including no fault signals, all of which were healthy dynamic strain signals of the bogie frame. They were used as historical healthy dynamic strain signals to construct a data set, and the data set was used to train the subsequent bogie frame fault warning model.

[0045] In step S2, the acquired signal data is standardized.

[0046] In step S3, a bogie frame fault warning model is constructed and trained. The model consists of two parts: a healthy dynamic strain data prediction model and a dynamic warning threshold generation module. The overall structure of the model is as follows: Figure 2 shown.

[0047] (1) Constructing a bogie frame fault warning model

[0048] (1.1) Construction of health dynamic strain data prediction model

[0049] Considering that the existing diagnostic methods for dynamic strain data usually use recurrent neural networks such as long-term and short-term neural networks to generate healthy data state space, and such methods have complex network structures and numerous parameters, the training process becomes quite time-consuming and complicated, which limits their application in real-time fault signal warning. On the contrary, these methods are more used for post-analysis to retrospectively diagnose the problem. Therefore, in order to achieve online fault warning, this embodiment selects a simple echo state network to generate a healthy data state space.

[0050] Specifically, the health dynamic strain data prediction model adopts the echo state network deep learning architecture. The topological structure of the echo state network consists of an input layer, a storage pool, and an output layer. Figure 3 As shown in the figure, the input layer and the output layer each contain K input nodes and L output nodes. The storage pool is a unique structure of the echo state network, which consists of N internal nodes and sparse node connection weights. The echo state network is used to predict the health data state. After the original dynamic strain signal of the bogie frame is input into the echo state network, the predicted value of the healthy dynamic strain data at the next moment can be predicted, that is:

[0051] y(t+1)=f(W out (E(t+1),u(t+1),y(t))) (2)

[0052] Where f(·) indicates that the activation function of the output node is the hyperbolic tangent function Sigmoid, y(t+1) and y(t) represent the predicted values ​​of the healthy dynamic strain at the t+1th step and the tth step, respectively, and W out is the output weight matrix, E(t+1) is the dynamic strain input of the bogie frame at the t+1th step, that is, a dynamic strain data composed of n dynamic strain sampling data from t+1-(n-1) to t+1, and u(t+1) is the storage pool state vector at the t+1th step.

[0053] In this embodiment, the trained healthy dynamic strain data prediction model is used to generate the predicted value of healthy dynamic strain data at the next moment, and the residual value of historical dynamic strain data is generated according to the predicted value of healthy dynamic strain data at the next moment and the actual dynamic strain data value.

[0054] (1.2) Construction of dynamic warning threshold generation module

[0055] like Figure 4 As shown in the figure, in the dynamic warning threshold generation module, a dynamic residual threshold generator based on extreme value theory is constructed for the dynamic strain data residual. Assume that x 1 ,x 2 ,…,x n is an independent residual sample vector with the same distribution F, where x jRepresents the jth dynamic strain residual value in a certain unit time. In order to determine the residual maximum distribution of these dynamic strain residuals, the optimal statistic of the tail distribution is established as:

[0056] M n =max(x 1 ,x 2 ,…,x n ) (3)

[0057] Where M n is the maximum value of the dynamic strain residual per unit time, and its probability distribution is:

[0058]

[0059] Because the F distribution is unknown, the approximate family G(x) of the model is used to estimate the extreme value, that is, the generalized extreme value distribution function G(x) is used to approximate the residual maximum value distribution of the residual value of the dynamic strain data, where the generalized extreme value distribution function is:

[0060]

[0061] Where μ(t), σ(t) and ξ represent the location parameter, scale parameter and shape parameter respectively, which are all solved and determined according to the maximum likelihood function. It can be expressed as:

[0062]

[0063] On the basis of the above-mentioned determination of the generalized extreme value distribution function, a confidence interval is established based on the function, and the confidence level p (i.e. Figure 4 p 1 ), that is, the confidence level p is established by combining the generalized extreme value distribution function with experience. When the dynamic strain data is input into the model, the dynamic residual warning threshold D at the next moment under the premise of confidence level p can be obtained. th , which can be expressed as:

[0064]

[0065] Furthermore, in order to ensure the accuracy of early warning, in addition to the dynamic threshold, this embodiment also designs two static thresholds based on the root mean square error RMSE and information entropy. Specifically, based on the residual value of historical health dynamic strain data, a static threshold based on the root mean square error and information entropy is constructed, where the information entropy static threshold H d The calculation formula is:

[0066]

[0067] In the above formula, N is the number of intervals of equal length, ni is the number of residual data falling within the equal-length interval i, T d Represents the total number of observed samples in the collection of N equal-length intervals, that is, the total number of samples.

[0068] The calculation of the root mean square error static threshold RMSE is:

[0069]

[0070] In the above formula, Δx is the residual of the dynamic strain value per unit time.

[0071] The above root mean square error RMSE and information entropy H d The thresholds are all static thresholds. During the training process, the confidence level p 2 It is calculated as 95% and does not change after training is completed.

[0072] Furthermore, different levels of fault warnings are performed according to the comparison results, namely: first, the dynamic strain data residual value is compared with the dynamic warning threshold. If the dynamic strain data residual value exceeds the dynamic residual threshold, a secondary warning signal is issued and the static threshold check comparison is triggered, otherwise no alarm is issued; secondly, the dynamic strain data residual value is compared with the static warning threshold. If the dynamic strain data residual value exceeds any static warning threshold, a primary warning signal is issued.

[0073] (2) Training the bogie frame fault warning model

[0074] (2.1) Training of health dynamic strain data prediction model

[0075] The training process of the bogie frame fault warning model does not need to directly rely on fault data, but only relies on healthy dynamic strain signal data to achieve excellent training results, greatly reducing the dependence on scarce fault data and effectively reducing the overall cost of model development. During the training process, unsupervised training is used for the healthy dynamic strain data prediction model, and the source domain health data set is input into the healthy dynamic strain number prediction model to construct the bogie frame health state space under different working conditions. In the prediction model, the state equation and output equation of the echo state network are:

[0076] u(t)=f(W in E(t)+Wx(t-1)) (11)

[0077] y(t)=W out u(t) (12)

[0078] Where f(·) represents the hyperbolic tangent function Sigmoid, E(t), u(t), and y(t) represent the input variable, state variable, and output variable of the storage pool at time t, respectively; W in∈R M×L , W∈R M×M , W out ∈R M Respectively represent the input weight matrix, the internal weight matrix of the reserve pool, and the output weight matrix, where the input weight matrix W in It is randomly generated in the initial stage and will not change after it is generated. is the actual dynamic strain value at time t.

[0079] In the training process of the health dynamic strain data prediction model, the output weight matrix W out Minimize, its expression is consistent with the cost function L(W out ,b) is:

[0080] W out =(U T U) -1 U T y (13)

[0081]

[0082] In the above formula, b is a constant bias term, is the actual dynamic strain value corresponding to the i-th moment, and ||·|| is the L2 norm of the calculation matrix. Among them, b plays the role of noise, avoiding the generation of singular results, and can effectively balance the error term and the complexity of the model, thus improving the network performance.

[0083] Based on the above output weight matrix, the cost function L(W out ,b), with the cost function L(W out ,b) is the objective function, and the training is iterated continuously by minimizing the objective function until the training is completed.

[0084] (2.2) Training of dynamic warning threshold generation module

[0085] For the dynamic threshold D in the threshold generation module th Based on the residual value of historical dynamic strain data, the generalized extreme value distribution function is used to approximate the residual maximum value distribution of the residual value of historical dynamic strain data. For the location parameter, scale parameter and shape parameter in the generalized extreme value distribution function, the maximum likelihood estimation function is constructed to solve. Specifically, when the cost function is minimized L(W out ,b) under the condition of maximum likelihood estimation to solve its optimal solution, its log-likelihood function l(μ,σ,ξ) is:

[0086]

[0087] Where m represents the number of dynamic strain measurement points per unit time.

[0088] Taking the above maximum likelihood estimation function as the objective function and the residual value of historical dynamic strain data as input, the training is iterated continuously by minimizing the objective function until the training is completed, and the determined values ​​of the position parameters, scale parameters and shape parameters are obtained to construct a dynamic warning threshold generation module that has completed the training.

[0089] In this embodiment, the healthy dynamic strain signal of the train operation history in the source domain is obtained, and the actual dynamic strain data in the source domain and the healthy dynamic strain data predicted by the neural network are firstly subtracted to generate residual data, and then the residual data is used as input to the dynamic warning threshold generation module. Through the above training process, the position parameters, scale parameters and shape parameters are preliminarily determined according to the historical data, and the dynamic warning threshold is preliminarily determined, that is, the dynamic warning threshold is generated at this time.

[0090] Furthermore, the dynamic warning threshold obtained by the above construction and training is an initial dynamic threshold determined and set in advance using historical data. In the actual operation process, the dynamic strain data residual of the target domain will also be input into the dynamic warning threshold generation module. With the input of the target domain data, both the position parameter and the scale parameter will continue to change, and then the dynamic warning threshold will be affected by the scale parameter and the position parameter and continue to change, so as to update the initial dynamic threshold and obtain the current dynamic warning threshold; then, the current dynamic strain data residual value is compared with the dynamic residual warning threshold to complete the warning of the dynamic strain data.

[0091] That is, finally, after constructing and training the bogie frame fault warning model, the ability to provide fault warning for the target domain bogie frame data is obtained. At this time, the target domain dynamic strain signal is data standardized and input into the bogie frame fault warning model. First, the healthy dynamic strain data prediction model in the model is used to output the predicted value of the healthy dynamic strain data at the next moment, and the dynamic strain data residual value is determined based on the actual dynamic strain data value and the predicted value of the healthy dynamic strain data at the next moment. Then, the dynamic strain data residual value is compared with the dynamic residual warning threshold and the static warning threshold, and the warning of the dynamic strain data can be completed through threshold judgment.

[0092] Embodiment 2

[0093] This embodiment provides a bogie structure fault early warning system based on dynamic strain signals, including:

[0094] The data acquisition and preprocessing module is used to acquire the real-time dynamic strain signal of the train operation and perform data preprocessing on the acquired signal;

[0095] The fault warning module is used to input the preprocessed real-time dynamic strain signal into the bogie structure fault warning model, firstly output the predicted value of healthy dynamic strain data at the next moment through the healthy dynamic strain data prediction model in the model, and determine the residual value of dynamic strain data based on the actual dynamic strain data value and the predicted value of healthy dynamic strain data at the next moment; then compare the residual value of dynamic strain data with the dynamic residual warning threshold and the static warning threshold, and perform fault warnings of different levels according to the comparison results;

[0096] Among them, the historical healthy dynamic strain signals of train operation are used to train the bogie structure fault warning model based on the healthy dynamic strain data prediction model and the dynamic warning threshold generation module.

[0097] Embodiment 3

[0098] This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps in the bogie structure fault warning method based on dynamic strain signals as described above are completed.

[0099] Embodiment 4

[0100] This embodiment also provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in the bogie structure fault warning method based on dynamic strain signals as described above are completed.

[0101] The steps involved in the above embodiments 2 to 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0102] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0103] The above description is only a preferred embodiment of the present invention. Although the specific implementation mode of the present invention is described in conjunction with the accompanying drawings, it is not a limitation of the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative work are still within the protection scope of the present invention.

Claims

1. A bogie frame fault early warning method based on dynamic strain signals, characterized in that: include: Acquire the real-time dynamic strain signal of the train operation and perform data preprocessing on the acquired signal; The preprocessed real-time dynamic strain signal is input into the bogie structure fault warning model. First, the healthy dynamic strain data prediction model in the model is used to output the predicted value of the healthy dynamic strain data at the next moment, and the residual value of the dynamic strain data is determined based on the actual dynamic strain data value at the next moment and the predicted value of the healthy dynamic strain data; Then the residual value of the dynamic strain data is compared with the dynamic residual warning threshold and the static warning threshold, and different levels of fault warning are performed according to the comparison results; Among them, the historical healthy dynamic strain signals of train operation are used to train the bogie structure fault warning model based on the healthy dynamic strain data prediction model and the dynamic warning threshold generation module.

2. A bogie frame fault early warning method based on dynamic strain signals as claimed in claim 1, characterized in that: According to the residual value of dynamic strain data, the dynamic residual warning threshold is determined through the dynamic warning threshold generation module, including: According to the input dynamic strain data residual value, the position parameter, scale parameter and shape parameter are updated to construct a generalized extreme value distribution function to approximate the residual maximum value distribution of the dynamic strain data residual value; Based on the generalized extreme value distribution function, a confidence level p is established, and the dynamic residual warning threshold at the next moment under the confidence level p is obtained.

3. The bogie structure fault early warning method based on dynamic strain signal according to claim 1, characterized in that: Different levels of fault warnings are given based on the comparison results, including: Compare the residual value of the dynamic strain data with the dynamic warning threshold. If the residual value of the dynamic strain data exceeds the dynamic residual threshold, a secondary warning signal is issued and a static threshold check comparison is triggered. Otherwise, no alarm is issued. The residual value of the dynamic strain data is compared with the static warning threshold. If the residual value of the dynamic strain data exceeds any static warning threshold, a first-level warning signal is issued.

4. The bogie structure fault early warning method based on dynamic strain signal according to claim 1, characterized in that: Determine static warning thresholds, including: Based on the residual values ​​of historical healthy dynamic strain data, a static threshold based on root mean square error and information entropy is constructed.

5. The bogie structure fault early warning method based on dynamic strain signal according to claim 1, characterized in that: The training process of the bogie frame fault warning model includes: Obtain historical healthy dynamic strain signals of train operation and perform data preprocessing on the acquired signals; Offline modeling is performed based on preprocessed historical healthy dynamic strain data, including: first, building a healthy dynamic strain data prediction model based on the echo state network, and using the historical healthy dynamic strain data for training; secondly, using the trained healthy dynamic strain data prediction model to generate the residual value of the historical dynamic strain data, and using the residual value of the historical dynamic strain data to build and train a dynamic warning threshold generation module based on extreme value theory.

6. A bogie frame fault early warning method based on dynamic strain signals as claimed in claim 5, characterized in that: The health dynamic strain data prediction model adopts an echo state network deep learning architecture, and the echo state network is used to predict the health dynamic strain data value, including an input layer, a storage pool and an output layer, wherein the input layer and the output layer each contain K input nodes and L output nodes, and the storage pool is composed of N internal nodes and sparse node connection weights; In the training process of the health dynamic strain data prediction model, a cost function is constructed based on the output weight matrix, and the cost function is used as the objective function. The training is continuously iterated by minimizing the objective function until the training is completed.

7. A bogie frame fault early warning method based on dynamic strain signals as claimed in claim 5, characterized in that: The training of the dynamic warning threshold generation module includes: Based on the residual values ​​of historical dynamic strain data, the generalized extreme value distribution function is used to approximate the residual maximum value distribution of the residual values ​​of historical dynamic strain data; For the location parameters, scale parameters and shape parameters in the generalized extreme value distribution function, a maximum likelihood estimation function is constructed for solution, including: taking the maximum likelihood estimation function as the objective function, taking the residual value of the historical dynamic strain data as the input, and performing continuous iterative training by minimizing the objective function until the training is completed, obtaining the determined values ​​of the location parameters, scale parameters and shape parameters, and constructing a dynamic warning threshold generation module after training.

8. A bogie structure fault early warning system based on dynamic strain signals, characterized in that: include: The data acquisition and preprocessing module is used to acquire the real-time dynamic strain signal of the train operation and perform data preprocessing on the acquired signal; The fault warning module is used to input the preprocessed real-time dynamic strain signal into the bogie structure fault warning model, firstly output the predicted value of healthy dynamic strain data at the next moment through the healthy dynamic strain data prediction model in the model, and determine the residual value of dynamic strain data based on the actual dynamic strain data value and the predicted value of healthy dynamic strain data at the next moment; Then the residual value of the dynamic strain data is compared with the dynamic residual warning threshold and the static warning threshold, and different levels of fault warning are performed according to the comparison results; Among them, the historical healthy dynamic strain signals of train operation are used to train the bogie structure fault warning model based on the healthy dynamic strain data prediction model and the dynamic warning threshold generation module.

9. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of a bogie structure fault warning method based on dynamic strain signals as described in any one of claims 1 to 7 are completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of a bogie structure fault warning method based on dynamic strain signals as described in any one of claims 1-7.

Images